Recommendation

Content recommendation on the internet refers to the use of algorithms and data analysis by online platforms to deliver personalized content to users based on their individual preferences, interests, and browsing behavior. Critically, recommendations are not in response to a preceding or historical user request - the platform offers them proactively, either in a user's primary view of the content of the site, or as a standalone "discovery" feature on the site.

For recommendations to be accurate, platforms have to invest in significant data mining, profiling, and predictive modeling efforts, so they're technically challenging for all but the largest online platforms to build well. But for the platforms that are able to make these investments and build this infrastructure, they have developed a hugely impactful tool: the capacity to turn existing user engagement into more user engagement - the key to unlocking a spiral of engagement. When platforms can offer relevant content to users before they explicitly seek it, or even know it is something they might want, they've inverted the relationship between user and platform: now the platform decides what to use the user's attention for, rather than the user deciding how they want to use the platform.

Where is content recommendation?

The most common places we see content recommendation are in the contexts of advertising and content discovery. In both of these worlds, the end goal is to increase platform engagement as measured through clicks, time on page, or other metrics of engagement, and because these goals are narrowly defined, the problem space is well adapted to machine learning techniques.

However, there are much less sophisticated, and much more ubiquitous uses of recommendation across the web. Recommended connections on social networks, autoplay at the end of a playlist, trending hashtags, each of these is an example where a platform is proactively suggesting content to a user in order to induce behavior or attention.